This project focuses on understanding how the brain constructs networks of interacting regions (i.e., neural networks) to perform cognitive tasks, especially those associated with language, and how these networks are altered in brain disorders. These issues are addressed by combining computational neuroscience techniques with functional neuroimaging data, obtained using PET or fMRI. The network analysis methods allow us to evaluate how brain operations differ between tasks, and between normal and patient populations. This research will allow us to ascertain which networks are dysfunctional, and the role neural plasticity plays in enabling compensatory behavior to occur. During the past year, criteria for assessing the adequacy of the goodness-of-fit of some of these network methods were examined (NeuroImage 11: 289-301, 2000), and a method for determining the importance of different network brain modules, for a given task, was developed (Hum. Brain Mapp. 9: 165-182, 2000). We have began delineating the functional networks involved in the production of spontaneous narrative speech in normal subjects using PET measurements of regional cerebral blood flow (rCBF), an index of neural activity. We determined the functional connectivity (evaluated as the correlation between rCBF in different brain areas) of left hemisphere (LH) brain regions. Our results demonstrate that LH perisylvian regions interact strongly with one another during language production, but not during a task requiring similar muscle movements and vocalizations as speech. In patients who stutter, many of these strong functional linkages seem to be absent, suggesting that LH language-production networks are abnormal during stuttering. We have also begun examining brain functional connectivity using fMRI. We investigated the functional connectivity of two areas of the left inferior frontal gyrus (LIFG) during orthographic processing of four kinds of visual stimuli - words, pseudowords, consonant strings and false fonts. We found strong functional connectivity between ventral LIFG and left posterior areas in occipital and temporal cortex only during the processing of semantically rich stimuli (i.e., words). Conversely, the posterior/dorsal part of LIFG had strong functional connections with left posterior occipital and temporal areas when the stimuli had phonological content (i.e., words, pseudowords and letter strings). These results support the notion that ventral LIFG is more specialized for semantic processing and that posterior/dorsal LIFG is more engaged in phonological processing. For language function, one important part of the LIFG is Broca's area, which is defined as the cytoarchitectonic areas 44 and 45 in the system devised by Brodmann. We used a probabilistic data set corresponding to cytoarchitectonically-defined Brodmann areas (BA) 44 and 45, which enables us to say for any given voxel in the stereotactic space of the Talairach atlas, what the probability is that this voxel is in BA44 or BA45. We applied this probabilistic atlas to PET data acquired during language production tasks (generation of narrative speech) from adults whose parents were deaf and who were fluent in both speech and American Sign Language (ASL). Narrative production was performed in separate PET scans using speech and sign. We found similar activations of the central parts of both Brodmann areas by both speech and sign, suggesting that in spite of the different modalities by which speech and sign are expressed, the same neural substrates in Broca's region are engaged. In order to understand the relationship between what is observed in functional neuroimaging studies and the underlying neural dynamics, we had previously constructed a large-scale computer model of neuronal dynamics that performs a visual object-matching task similar to those designed for PET studies (Cerebral Cortex 8:310-20, 1998). The model is composed of elements that correspond to neuronal assemblies in cerebral cortex, and contain different elements that are based on types identified by electrophysiological recordings from monkeys as they perform similar tasks. It includes an "active" memory network involving the occipitotemporal visual pathway and a frontal circuit, and is capable of performing a match-to-sample task in which a response is made if the second stimulus matches the first. A PET study is simulated by presenting pairs of stimuli to an area of the model that represents the lateral geniculate nucleus. Simulated PET data are computed from the model as it performs the tasks by integrating synaptic activity within the different areas. Simulated PET data similar to that found in actual PET delayed match to sample visual tasks were obtained, as were the correct neuronal dynamics in each brain region. In the past year, we expanded the model to simulate fMRI studies, particularly those that use an event-related design (Hum. Brain Mapp. 8: 137-142, 1999). It was found that cognitive activity not under experimental control, especially in anterior brain regions (e.g., frontal cortex) can confound interpretation of fMRI activity in terms of its underlying neural substrate. We have also investigated how neural inhibition is manifested in PET/fMRI signals. We found that neuronal inhibition can cause measured PET/fMRI activity to either increase or decrease, depending on the amount of excitatory activity present. We have also begun an expansion the model so that it can also simulate auditory processing. We constructed a large-scale neural model of the auditory pattern recognition ('what') pathway in cortex consisting of 4 regions: A1 (primary auditory cortex), A2 (secondary auditory cortex), STG (superior temporal gyrus) and PFC (prefrontal cortex). There are different scales of temporal integration in the first 3 stages of the model, with A1 having the smallest temporal window and the STG the longest. Compared to a control task consisting of steady-state tones, transient tonal patterns resulted in increased simulated rCBF in all regions, with the highest increases occurring in the STG and PFC regions, suggesting that the complexity of the stimulus is correlated with greater activation in higher auditory processing areas. This modeling effort will be combined with PET and fMRI experiments of the same kind of tasks used in the simulation studies. We also have incorporated into the large-scale model a way to simulate transcranial magnetic stimulation (TMS). During TMS, a strong, changing magnetic field applied to a region on the scalp induces intracranial electrical currents that can alter regional neuronal function. TMS exerts both excitatory and inhibitory effects on stimulated neural tissue, although little is known about the exact neurobiological mechanisms by which TMS alters neuronal function. TMS has been used in conjunction with PET to examine interregional connectivity of human cerebral cortex. In the present simulations, TMS was applied to the large-scale neural model of the visual processing pathway to investigate its effects on rCBF in regions connected to the stimulated area.We showed that we could reproduce the experimental finding of decreased rCBF following TMS in brain regions functionally connected to the stimulation site, but only if TMS was applied to the inhibitory neuronal population.